Direct Mail Still Works: AI for List Selection, Segmentation, and Copy Testing
Every few years someone announces that direct mail is finished, and every year a substantial share of unrestricted revenue at mass-market and mid-level nonprofits arrives in a reply envelope. The channel is not dying. It is expensive, unglamorous, and governed by a set of rules most digital-native fundraisers were never taught, which is a different problem. This is a working guide to the parts that determine whether a mailing makes money: who you mail, how you divide them, what goes in the envelope, how you test it, and how you count the result without fooling yourself.
Direct mail occupies a strange position in nonprofit fundraising. It is simultaneously the channel most likely to be described as obsolete in a strategy meeting and the channel that pays for a large share of the general operating budget. The tension is not really about the medium. It is that mail costs money before it earns any, which makes every decision visible and every mistake expensive, while a badly targeted email costs almost nothing and disappears quietly.
The scale is worth stating plainly. According to the M+R Benchmarks 2026 study, participating nonprofits received an average of $0.66 in direct mail revenue for every $1 received online, direct mail revenue rose 9 percent in 2025, and the average direct mail gift was $120, with smaller organizations averaging $163. For an organization whose donor base skews older, mail is not a supplement to the digital program. It is a co-equal revenue engine that happens to have a longer production cycle.
What has changed is not whether mail works but how much slack the economics allow. Postage rises, paper and lettershop costs rise, and response rates on cold acquisition have been under pressure for years. The organizations still doing well in the channel are not the ones with the most moving copy. They are the ones making better decisions about which names go into the file, and that is a data problem before it is a creative problem.
This is where AI earns its place. Large language models are genuinely useful for interrogating a donor file in plain English, for drafting the eight copy variants nobody has time to write, and for reading test results without the motivated reasoning that creeps into a review of your own package. They are not useful for deciding what your organization sounds like, and they cannot rescue a file that has not been cleaned. What follows separates the two.
Why Mail Still Performs, and Why the Numbers Mislead
A mailed appeal has physical advantages that no amount of inbox optimization replicates. It arrives without a spam filter between it and the recipient, it sits on a kitchen counter for days rather than scrolling out of view in ninety seconds, and it carries a reply device that turns intent into a completed gift without requiring anyone to remember a password. For donors in their seventies and eighties, many of whom joined the file through mail decades ago, it is also the format they trust most. That cohort is disproportionately responsible for the largest annual gifts outside of major giving, and telling them to use a donation form is not a strategy, it is an eviction notice.
The comparison that gets quoted, mail's response rate against email's, is technically true and analytically useless. Mail response is measured against a list you paid to reach, while email response is measured against a list that cost almost nothing to send to. A 4 percent response on a housefile mailing and a 0.2 percent response on an email blast are not competing figures. They are figures from two different cost structures, and the only honest comparison is net revenue per dollar spent within each channel, at each audience segment, over a period long enough to capture repeat giving.
The more interesting effect is what happens when the two run together. Mail drops reliably produce a lift in online giving during the following two to three weeks, because a meaningful share of recipients read the letter, put it down, and give at a website or through a phone later. If your attribution treats those gifts as organic web traffic, the mailing looks worse than it is and the online program looks better than it is. Over several years of that accounting, organizations cut mail budgets based on numbers that were never measuring mail.
The practical fix is not perfect attribution, which does not exist in this channel. It is a set of deliberate approximations you apply consistently: a unique URL or short vanity domain on the reply device, a source-coded landing page, a matchback of online gifts against the mail file by name and address for the weeks following each drop, and a baseline period so you can see what online giving looks like when no mail is in home. None of these is exact. Applied the same way every time, they are stable enough to compare one mailing to another, which is the actual decision you need to make.
What a mail drop is really producing
Four revenue effects, only one of which lands in the reply envelope
- Checks returned in the reply envelope, coded to the mailing and easy to count
- Online gifts in the two to three weeks after the drop, often uncoded and misattributed
- Phone gifts from donors who call the number on the letter rather than write a check
- Renewal and retention effects that show up in later mailings, not this one
The Economics You Need Before You Mail a Single Piece
Mail is arithmetic. Before any creative conversation, you should be able to state your fully loaded cost per piece, your break-even response rate at your expected average gift, and your intended treatment of any segment that will not break even on the first mailing. Organizations that cannot state those three numbers are not running a direct mail program, they are running a series of hopeful drops.
Cost per piece is the sum of creative, printing, list acquisition or rental, data processing, lettershop, and postage. Postage is the piece nonprofits have the most leverage over, because 501(c)(3) organizations can apply for authorization to mail at Nonprofit USPS Marketing Mail prices, which run roughly half of commercial Marketing Mail rates. That authorization is not automatic. You apply with PS Form 3624 at the Post Office where your mailings will be deposited, with your formative papers and IRS exemption letter attached, and once approved the authorization applies nationwide. Organizations that have never filed it are paying double on every drop for the want of a form.
Within nonprofit rates there is a second tier of savings that depends entirely on data quality. Presort and automation prices require that your addresses be standardized, barcoded, and deliverable, so the difference between a mixed-tier rate and a five-digit automation rate is a function of how well your file was processed, not how well your letter was written. This is the first place where data hygiene stops being a tidiness concern and becomes a line item.
Break-even response rate is then straightforward. Divide your cost per piece by your expected average gift. At a fully loaded cost of $0.55 per piece and an expected average gift of $45, you need roughly 1.2 percent response to break even before considering any downstream value. Run that calculation for every segment separately, because a lapsed segment with a $28 average gift and an active segment with a $95 average gift have entirely different break-even points against the same cost per piece.
Acquisition is the part that confuses boards, and it needs to be explained before the first invoice lands rather than after. Cold acquisition mail usually loses money on the first gift, and it is supposed to. The M+R data puts the average return at $0.41 per dollar spent on prospect audiences, against $0.90 on lapsed audiences and $4.51 on active donors. Acquisition is a deliberate investment in a donor who will be worth several times the acquisition cost over the following years, recovered through renewal, upgrade, monthly conversion, and eventually planned gifts. Judged on first-gift revenue alone, every acquisition program in the country is a failure. Judged on lifetime value, a well-run one is the reason the organization still has a donor file at all.
That argument only holds if you actually measure lifetime value and actually run the retention program that produces it. An organization that acquires donors at a loss and then does nothing to keep them is simply losing money on a schedule. The economics of acquisition depend on what happens in months two through thirty-six, which is a question about your donor lifecycle rather than about the mailing.
Numbers to fix before creative starts
Per segment, not per mailing
- Fully loaded cost per piece, including data processing and list costs
- Expected average gift based on that segment's actual history
- Break-even response rate, and the response rate you actually expect
- Net revenue per piece, which is the number that decides quantity
- For acquisition, the payback horizon and the retention plan behind it
Cost mistakes that repeat every year
Usually invisible until the reconciliation
- Never filing for nonprofit authorization and paying commercial rates
- Losing automation discounts because the file was not properly processed
- Mailing deep into unproductive segments to hit a round quantity
- Charging acquisition losses against the annual fund without a payback model
- Counting gross revenue in the board report and net revenue nowhere
List Selection and Segmentation: Where the Money Actually Is
If you improve one thing about your mail program this year, improve the selection. Creative differences move response by fractions of a percentage point. Selection differences move it by multiples. The gap between mailing your whole file and mailing the right two thirds of it is frequently the difference between a program that funds operations and one that consumes them.
The foundational division is housefile against acquisition. Your housefile consists of people who have given before, and it subdivides into active donors, recently lapsed donors, deeply lapsed donors, and non-donor constituents such as event attendees, volunteers, and newsletter subscribers. Acquisition is everyone else, reached through rented lists, list exchanges, or cooperative database models. These two halves of the program have different costs, different response rates, different creative requirements, and different success criteria, and running them on one set of assumptions is the most common structural error in the channel.
Within the housefile, recency, frequency, and monetary value remains the workhorse framework, and it remains so because it keeps outperforming more elaborate alternatives. Recency is the strongest single predictor of whether someone gives again: a donor who gave three months ago is far more likely to respond than one who gave three years ago, regardless of how large the older gift was. Frequency captures habit, which is why a donor with five gifts of $25 is usually a better prospect than one with a single gift of $125. Monetary value sets the ask string. Combine them into a grid and you get segments you can price, treat, and measure separately.
Lapsed reactivation deserves its own windows rather than a single lapsed bucket. Donors who last gave thirteen to twenty-four months ago behave very differently from those at thirty-seven to sixty months, and both behave differently from a donor who gave once, years ago, in response to a disaster appeal. Mail the recent lapsed group with a renewal posture that assumes the relationship continues. Mail the deeply lapsed group less often, with reacquisition creative that reintroduces the organization, and be willing to stop entirely at the point where the segment stops paying for itself. The full mechanics of these windows are covered in our guide to lapsed donor reactivation.
For acquisition, the three sources behave differently enough to be worth naming. List rental gives you one-time use of another organization's donors or a commercial file, usually at a per-thousand price, with the rental terms enforced by seed names planted in the file. List exchange trades your donor names against another nonprofit's on a roughly equal basis, which costs less cash but means your donors are being mailed by someone else. Cooperative databases pool transaction data from many participating organizations and build models to identify prospects who look like your best donors, which typically outperforms flat rentals because the selection is modeled rather than categorical. Co-ops require you to contribute your own file, and whether that trade is acceptable is a policy decision your board should make knowingly rather than one a consultant makes in a production schedule.
Suppression is the unglamorous half of selection and the half that protects your reputation. Before any file goes to the lettershop it should be run against deceased records, against do-not-mail and mail preference registries, against your own internal do-not-solicit flags, and against current major donor and planned giving prospect lists so that a personally cultivated donor does not receive a generic $25 ask from the mass program. Mailing a household that reported a death months ago is the kind of error donors remember for a decade, and the suppression pass that prevents it costs a fraction of a cent per record.
Segments worth separating
Each gets its own ask string and frequency
- Active multi-gift donors, split by giving level
- First-time donors inside their first twelve months, where attrition is steepest
- Recently lapsed at thirteen to twenty-four months, treated as renewal
- Deeply lapsed beyond three years, treated as reacquisition and mailed sparingly
- Monthly donors, who usually need a different appeal or none at all
- Non-donor constituents such as volunteers and event attendees
Suppression passes before every drop
Cheap to run, expensive to skip
- Deceased records, including registrations made by family members
- Mail preference and do-not-solicit registries for prospect mail
- Your own internal opt-outs, honored across every channel not just mail
- Major gift, planned giving, and board prospect lists handled personally
- Recent complainants and anyone who asked to be mailed once a year only
Where AI Genuinely Helps With Segments
The bottleneck in segmentation is rarely analytical sophistication. It is that the person who understands the donors cannot write the queries and the person who can write the queries does not know which questions matter. A language model connected to an export of your giving history collapses that gap. You describe the segment in plain English, it produces the logic, and you interrogate the result rather than waiting three weeks for a report.
Useful questions look like this. Which donors gave in each of the last three years but have not given in the current one. Which first-year donors upgraded after their second gift, and what did their first gift amount have in common. Which segments produced net revenue below zero in the last four mailings. Which addresses appear more than once under slightly different names. What is the average gift by recency band, and where does it fall off a cliff. These are all answerable from data you already hold, and the value is in asking twenty of them in an afternoon instead of three of them in a quarter.
AI also helps with the ask string, which is where a surprising amount of revenue hides. Given a donor's giving history, a model can produce segment-level ask arrays anchored on highest previous contribution rather than on one organization-wide default, and it can flag the segments where your current defaults are asking well below what the donors have already demonstrated they will give. This is adjacent to formal propensity modeling, which our guide to donor scoring models treats in more depth, but it does not require a data science function to start.
Two cautions. First, a model given a spreadsheet will answer confidently whether or not the underlying data supports the answer, so treat every output as a hypothesis to verify against the source rather than a finding. Second, do not paste donor-level detail into a consumer tool without reading the vendor's data handling terms and checking it against your own privacy commitments. De-identified or aggregated extracts answer most segmentation questions perfectly well, and our guide to donor data privacy with AI tools covers where the lines sit.
The Package: Drafting Copy With AI Without Losing Your Voice
A direct mail package is not a letter. It is an outer envelope, a letter, a reply device, a reply envelope, and often a lift note or an insert, and each component does a specific job in a specific order. The outer envelope has one job, which is to get opened. The letter has one job, which is to get read far enough to produce a decision. The reply device has one job, which is to make the gift easy and to carry the coding that lets you measure everything else. Writing the letter beautifully and treating the other components as afterthoughts is the most common creative failure in the channel.
This component structure is what makes AI drafting genuinely useful here rather than merely fast. Producing eight outer envelope teaser lines, four opening paragraphs with different emotional entry points, three reply device headlines, and two lift note approaches is a volume problem, and volume is what these tools are for. A human writer produces two teaser options because producing eight is tedious. A model produces eight in a minute, most of which are discardable, which is exactly how creative ideation is supposed to work.
Voice is the part you have to protect deliberately. A letter signed by the executive director should sound like the executive director, and generic model output sounds like nobody. The technique that works is grounding rather than instruction: give the model three or four of your best-performing past letters, ask it to describe the voice it observes in specific terms such as sentence length, use of the first person, degree of directness in the ask, and typical opening move, then have it draft against that description. Correcting a draft that is 80 percent right is a fundamentally different task from correcting one written in the default register of a chatbot. The broader question of how machine-assisted appeals compare to human-written ones is worth reading alongside this, and we covered it in our piece on AI-generated versus human fundraising appeals.
Personalization in mail should go past the salutation, and AI makes the marginal cost of doing so much lower. Referencing a donor's previous gift amount, the year they first gave, the specific program they have supported, or their membership anniversary produces measurable lift, and variable data printing has made the production side routine. The constraint is data accuracy, not printing. A letter thanking someone for eleven years of support when they gave twice in 2019 does more damage than no personalization at all, which means the personalization fields you use should be the ones your file supports rather than the ones the printer can technically execute.
Finally, write for the eyes of the people actually reading it. A donor base skewing toward seventy-five and older is reading in imperfect light, possibly without glasses to hand. That argues for serif body text at twelve to fourteen points, generous leading, short paragraphs, real margins, dark ink on light stock rather than any reverse treatment, and an unambiguous ask repeated in the postscript. Asking a model to check a draft for reading level, sentence length, and unexplained jargon is a thirty-second step that catches more problems than it should, because the people writing the letter are not the people struggling to read it.
What each component has to accomplish
Draft them as a set, not as a letter plus accessories
- Outer envelope: earn the open, without overpromising what is inside
- Letter: one story, one ask, an explicit deadline or reason to act now
- Postscript: restate the ask, because it is the second thing most people read
- Reply device: ask array, source code, prefilled donor details, online option
- Lift note: a second voice, often a beneficiary or board member, adding proof
- Reply envelope: postage-paid where the segment economics justify it
Test Discipline: The Part AI Cannot Shortcut
The arrival of tools that generate twenty copy variants in a minute has created a specific temptation, which is to assume the model can also tell you which variant will win. It cannot. A model can tell you which draft reads more clearly and which follows direct response conventions more faithfully. Whether your donors will give more money in response to it is an empirical question about a particular audience at a particular moment, and the only instrument that answers it is a live test.
The structure is unchanged from long before any of this. You have a control, which is the package that has performed best to date, and you have a challenger that differs from it in one respect. You mail both to randomly assigned, statistically equivalent portions of the same segment, in the same drop, into the same mail stream. If the challenger wins by enough to matter, it becomes the new control and you build the next challenger against it. That cycle, run twice a year for five years, compounds into a package that dramatically outperforms whatever you started with, and it is why organizations with long-running mail programs have controls that seem impossible to beat.
One variable at a time is the rule people break most often, usually under budget pressure. If the challenger changes the teaser, the opening, the ask string, and the signer simultaneously and it wins by 12 percent, you have learned that this particular combination beat the control, and nothing at all about why. Next year, when you want to change the signer, you are back to guessing. Bundled tests are not useless, they are just not learning, and the distinction matters in a channel where the learning is the asset.
Quantity is the other discipline. A test on 2,000 names at a 1 percent response rate produces twenty gifts, and twenty gifts cannot distinguish a real difference from noise. As a working rule, size each test cell so that you expect at least a hundred responses, and be honest when your file cannot support that. Small organizations often cannot test creative meaningfully, which is fine and worth admitting. They should test selection instead, where the effects are larger and the quantities needed are smaller, or adopt proven conventions rather than burning their file on inconclusive experiments. This is the same statistical logic that governs testing in digital channels, which our guide to subject line testing approaches from the email side.
Seed names round out the practice. Plant a handful of addresses you control, including staff, board members, and a post office box, into every drop so you know when mail actually landed. In-home dates are estimates, and the difference between mail arriving the week before a holiday and the week after it is large enough to invalidate a comparison between two drops. Seeds also verify that rented lists are being used within their terms, and they catch production errors such as a segment receiving the wrong package, which is the kind of thing that is obvious in a mailbox and invisible in a report.
A test that will actually teach you something
Everything else is a more expensive way to guess
- A defined control, with its historical performance written down
- One variable changed, chosen because you expect it to matter
- Random assignment within the same segment, not one segment against another
- Cell sizes large enough to expect roughly a hundred responses each
- Same drop date, same mail stream, seeds in every cell
- A decision rule agreed in advance about what counts as a win
Production and Data Hygiene, Where Mailings Quietly Fail
Nothing about this section is interesting and all of it decides whether the mailing works. Mail is the only fundraising channel where a data error becomes a physical object sitting in someone's hand, and where a postal rule you did not know about becomes a five-figure surcharge at the acceptance unit.
Address standardization comes first. Coding Accuracy Support System processing corrects and standardizes addresses against postal records, appends the ZIP+4 and delivery point information, and is the prerequisite for the barcoding that earns automation discounts. Separately, the Move Update standard requires mailers claiming presort or automation prices to have matched their addresses against an approved change-of-address source, with the commonly cited threshold being at least 95 percent of addresses updated within 95 days of the mail date. Falling short does not usually produce a polite warning. It produces denial of the discounted rates at verification, on the whole mailing.
Merge and purge sits alongside that work and handles a different problem: the same household appearing several times across your housefile and every rented list in the drop. Good merge and purge collapses duplicates, applies a priority order so that housefile records survive against rented records, and reports back which rented list supplied each surviving name so you can evaluate list performance afterwards. It also spares a donor the experience of receiving four identical letters, which they will mention, and which will be the thing they remember about the campaign.
Source coding is the piece most likely to be handled carelessly and most likely to be regretted. Every cell in every drop needs a unique code that is printed on the reply device, keyed at gift entry, and stored in the CRM in a field that reporting can actually reach. The coding scheme should identify the campaign, the drop, the segment, the package version, and the list source, and it should be designed once rather than improvised per mailing. An organization with clean source codes can answer any question about its mail program in an afternoon. An organization without them is permanently guessing, and no amount of analysis, AI-assisted or otherwise, recovers information that was never captured. If your file is already in poor condition, that is a prerequisite rather than a parallel project, and our guide to cleaning up a nonprofit CRM is the place to start.
Mail dates deserve more respect than they usually get. Nonprofit Marketing Mail is not expedited, delivery windows vary by entry point and season, and a drop scheduled to land the second week of December can easily arrive during the week between the holidays when nobody opens anything. Plan backwards from the in-home window you want, add real buffer for data processing and lettershop time, and confirm the entry plan with your mail house rather than assuming. The most carefully tested package in the world underperforms if it arrives at the wrong time.
Measuring Results Honestly
Four numbers describe a mailing, and reporting fewer than four is how programs drift. Response rate tells you whether the selection and the offer worked. Average gift tells you whether the ask string was right. Net revenue per piece combines both against cost and is the number that should drive next year's quantities. Cost to raise a dollar puts the mailing in context against your other channels, provided you compute it the same way everywhere.
Report all four by segment, not just in total. A mailing that returned $1.40 per dollar spent overall may contain an active donor segment returning $4.00 and an acquisition segment returning $0.35, and the aggregate figure conceals the only decision the report exists to inform, which is what to do differently next time. Segment-level reporting is also what makes the board conversation about acquisition losses tractable, because the loss is visible as a deliberate investment in a named segment rather than as a drag on an otherwise healthy total.
Then set a reporting window and keep it. Mail responses arrive over weeks, with a long tail that can continue for months, so a figure pulled at fourteen days and compared against one pulled at ninety days is not a comparison. Pick a standard horizon, commonly sixty or ninety days from in-home, report every mailing at that horizon, and note the tail separately if you want it. Consistency matters more than which horizon you choose.
Capture the online gifts that the mailing caused. Run a matchback of online donors against the mailed file for the weeks after each drop, compare online revenue against a baseline period with no mail in home, and use a distinct landing page or vanity URL on the reply device. Expect to undercount regardless, because some donors will find the website unaided and some will give months later. Undercounting consistently is still far better than attributing those gifts to a channel that did not produce them.
Finally, measure multi-year value, because the single-mailing view systematically undervalues the program. Track cohorts by acquisition source and watch what they give across the following three years, including second gift rate, upgrade rate, monthly conversion, and eventual planned gift inquiries. A rented list that produces expensive first gifts but retains well is a better investment than one that produces cheap first gifts and evaporates, and you cannot see the difference inside a single campaign report. This is also where mail earns its most underrated return, since a large share of legacy commitments come from long-tenured small-gift donors, a pattern our article on legacy giving explores in more depth.
The reporting pack for every drop
Same fields, same window, every time
- Quantity mailed, cost per piece, and total cost by segment
- Response rate and average gift by segment and package version
- Net revenue per piece and cost to raise a dollar
- Attributed online and phone gifts, with the method stated
- Test cell results with the decision taken and the reason
- Cohort performance for acquisition sources mailed in prior years
What AI Cannot Do Here, and the Privacy Line on Lists
AI does not choose your mission voice. The register your organization uses when asking for money is a strategic decision reflecting who you are and who your donors believe you to be, and it is made by people who carry the relationship. A model can imitate a voice you have already established and it can maintain consistency across a large volume of copy, which is genuinely valuable. It cannot decide whether your appeals should be urgent or measured, whether beneficiaries are named or protected, or how close to the edge your imagery sits. Those are questions about your mission's integrity, and they are covered more broadly in our piece on transparency in AI-assisted fundraising.
AI does not substitute for a clean file. If addresses are stale, duplicates are rampant, source codes were never captured, and gift history is incomplete, then every segmentation query returns confident answers built on bad inputs. Models are especially poor at flagging their own missing context, so an analysis of a file that silently excluded a third of gifts reads exactly like an analysis of a complete one. Getting the file right is prior work, not parallel work, and it is treated directly in our guide to putting clean data first.
AI also does not fix a bad offer. If the ask is vague, if the case for support does not explain what the money does, or if the organization has not communicated with these donors in eighteen months, the package underperforms regardless of how polished the prose is. Direct response rewards clarity about a specific need and a specific amount, and no amount of generated variation substitutes for having something worth asking about. This is the same discipline the best mid-level donor programs apply when they move donors up from the mass file.
On privacy, mail occupies an odd legal position in the United States. Most state consumer privacy laws, including California's, largely exempt nonprofit organizations, and the industry-run mail preference and deceased suppression registries are voluntary industry tools rather than statutory rights. The practical consequence is that what you do with donor names is governed mostly by your own commitments rather than by a regulator, which raises the stakes on those commitments rather than lowering them.
So be specific with donors and be specific with yourself. If you rent or exchange your donor names, say so plainly in your privacy policy, offer an easy way to opt out of that use while remaining a donor, and honor it permanently across every future exchange. If you do not exchange names, say that too, because it is a genuine trust asset worth stating. Be deliberate about contributing your file to a cooperative database, since that is a form of sharing even when the co-op never discloses individual records back. And apply the same care to what you feed an AI tool: donor names and addresses are personal information regardless of which legal regime does or does not cover them, and a donor who learns their giving history was pasted into a consumer chatbot will not be reassured by an exemption.
Keep these with people
Judgment, relationships, and accountability
- The mission voice and how far the emotional appeal goes
- Consent decisions about renting, exchanging, or pooling donor names
- Whether a segment gets mailed at all, when the economics are marginal
- Final sign-off on every piece that carries a leader's signature
- Reading a test result that contradicts a preference you hold strongly
Conclusion
Direct mail is not a nostalgic holdover. It is a disciplined, measurable channel that continues to deliver real unrestricted revenue, particularly from the older donors who fund a great deal of nonprofit operating work and who are also the most likely source of future legacy gifts. What it demands is precision about economics and selection, which is exactly what tends to erode when an organization treats mail as a legacy obligation rather than as a program someone actively runs.
AI is a real advantage in that program, and the advantage is concentrated in specific places. It lets a fundraiser interrogate a donor file in plain language instead of waiting on a query queue. It produces the volume of copy variants that proper testing requires and that nobody has time to write by hand. It drafts the segment-level ask arrays, the reporting summaries, and the readability checks that get skipped. It surfaces patterns across several years of results that are invisible inside a single campaign report.
What has not changed is that the file decides the outcome, the test decides the package, and the people decide the voice. Choose the right names, suppress the ones that should not be mailed, segment by behavior rather than by habit, change one thing at a time, and count what happened honestly including the gifts that arrived online. Do those things and mail will keep paying for itself and for a good deal else besides. Skip them, and no tool available now or later will make the difference.
Get More Out of Every Drop
We help nonprofits sharpen list selection, build segments they can defend, and use AI where it actually improves net revenue per piece.